Hi,
I am doing a binomial GLMM with a random intercept using the formula below,
but I always get the same warning message.
> m01 <- lmer(pres~ HT + DN + dtree + DNm + cmnhi + cmxes + cplan + craan
+
lfphal0100 + lfov0100 + lfop0100 + (1|plot), family=binomial, data=vphal,
verbose=TRUE)
0: 6309.9448: 0.459924 -5.20747 -0.378722 0.558779 -0.200922
-0.0488451 -0.397844 0.367916 -2.09820 0.233011 0.489537 0.156096 0.233738
1: 5328.6489: 1.44553 -5.30486 -0.331482 0.584113 -0.178051
-0.0525247 -0.374922 0.323662 -2.00877 0.204221 0.517675 0.123727 0.183290
2: 5102.8094: 1.90302 -6.00849 -0.122106 0.800716 -0.0894924
-0.161499 -0.336434 0.351752 -1.73945 0.117255 0.419783 0.428155 0.205801
3: 5047.7542: 1.97542 -6.03480 -0.190565 0.764629 -0.139488 -0.185599
-0.353159 0.435037 -1.79551 0.139551 0.427923 0.436322 0.198042
...
63: 4869.7669: 2.38299 -6.96339 -0.272766 0.911792 -0.340779 -0.295437
-0.436908 0.777541 -1.94732 0.191853 0.419920 0.514350 0.165871
64: 4869.7669: 2.38299 -6.96339 -0.272766 0.911792 -0.340779 -0.295437
-0.436908 0.777541 -1.94732 0.191853 0.419920 0.514350 0.165871
65: 4869.7669: 2.38299 -6.96339 -0.272766 0.911792 -0.340779 -0.295437
-0.436908 0.777541 -1.94732 0.191853 0.419920 0.514350 0.165871
66: 4869.7669: 2.38299 -6.96339 -0.272766 0.911792 -0.340779 -0.295436
-0.436908 0.777541 -1.94732 0.191853 0.419920 0.514350 0.165871
67: 4869.7669: 2.38299 -6.96339 -0.272766 0.911792 -0.340779 -0.295436
-0.436908 0.777541 -1.94732 0.191853 0.419920 0.514350 0.165871
Warning message:
In mer_finalize(ans) : false convergence (8)>
I read the various posts in the R-help list about it, and followed their
advice. I standardized the fixed factors and also used glmer with nAGQ, but
I still get the same warning message. Is it possible to get this error when
the number of zeros in the response variable is very high? I have 30281
observations grouped in 2402 plots (=random factor); 1299 observations from
the 30281 correspond to presences, and the rest to absences. Which is the
most appropriate way to overcome this problem?
Do you know of any pdf or book providing a detailed and friendly description
of the use of lmer and glmer functions, and the use of LA or nAGQ
approximations in GLMM? I read Zuur et al (2009), but they do not provide
any recommendation about the use of LA vs nAGQ under different conditions.
Thank you very much in advance.
Núria
[[alternative HTML version deleted]]